Support executing LSP commands

[Feel-ix-343]

Check for existing issues

• Completed

Describe the feature

Some language servers use the LSP command interface to provide some more custom functionality.

Popular editors such as vscode and neovim support this.

If applicable, add mockups / screenshots to help present your vision of the feature

I find how helix implements this (though incomplete) to be pretty great UX. Here is a gif of me using LSP commands with my language server markdown-oxide to open some daily notes.

[rcjsuen]

interface ClientCapabilities {
	workspace?: {
		/**
		 * The client supports applying batch edits
		 * to the workspace by supporting the request
		 * 'workspace/applyEdit'
		 */
		applyEdit?: boolean;
	}
}

From what I can tell here, applyEdit is not currently declared as a capability. Some language servers will "execute" the requested commands by actually delegating back to the client to apply a WorkspaceEdit (by sending a workspace/applyEdit request to the client) so Zed should take a look at the workspace/applyEdit request also when looking at workspace/executeCommand.

[kelveden]

As a reference, here is the specification for execute command.

From a selfish perspective, I'd love to see this feature added to enable us Clojure engineers to leverage some of the s-expression specific "power tools" from clojure-lsp - as you can see from that matrix, there are a bunch of those commands that aren't exposed as code actions.

[AlbertMarashi]

Need this to close some SvelteKit missing type issues

• Svelte types not automatically inferred in .ts & .js files #4577

[lnay]

There is currently no way (as far as I am aware) to send a textDocument/build or textDocument/forwardSearch to the language server. These are non-standard requests implemented by texlab which are important for LaTeX and the lack of these pushes users to configure bespoke tasks tailored to their setup (rzukic/zed-latex#29).

Ideally users would have these requests associated to a keybinding.

I envision a solution where the zed_extension_api exposes a way to send these requests, and allow them to register actions which can send these requests, and be bound by the user to a key combination. The internal Rust extension seems to have the ability to directly register tasks, however they can only run executables.

[AlbertMarashi]

+1 on that. I also wanted an ability for extensions to expose actions that use some of the underlying LSP features

[lnay]

Maybe tag "extension infrastructure"?